Depth Appearance From LSS Multiview Backbone

Hi, I've been trying to utilize your implementation of multiview LSS backbone for my thesis. I was wondering if the returned depth should be visually interpretable. I am aware that the depth tensor is spatially downsampled from original input images, but still I expect to get coarse depth image. Right now i just get a full noise like image(see image below) after decoding depth bins to meters (argmax(depths) * bin_size). I was expecting to see something like depth net output vis here

Start with the multiview LSS backbone's returned depth tensor and the reported decoding expression, argmax(depths) * bin_size; compare its behavior with the depth-net visualization in the MatrixVT README linked in the issue. Done means determining whether the decoded result should be a coarse visual depth image and documenting or correcting the cause of the reported noise.